What Deel knows about your staff, and what the connector keeps
Deel is where a company runs payroll and the administration of the people it employs. A person's record there carries their job title and their department, and also their pay, their date of birth, their personal email address, their phone numbers and their identity document.
The connector keeps four fields and stops there. The full name, the job title, the department as written on the record, and the name of the direct manager. Those four are enough for an AI to know who does what and who reports to whom, which is exactly the question an org chart answers.
So an AI plugged into your Kastel can say who handles purchasing, who signs off a decision, which department owns a given subject. It can say nothing about a salary, because the field was never read. The other sources your Kastel can read are gathered in the connector catalogue.
Deel's API returns the whole record, and the connector throws most of it away
Deel's interface cannot return part of a record. When the connector asks for the directory, it receives complete records, pay included. Some HR tools accept a request for a handful of fields only, and Lucca is one of them; Deel's people list does not offer that.
The decision is therefore taken at the point of writing. The connector picks the four fields it has named off the record it received, writes that text, and lets the rest go. The complete record Deel returned is written nowhere, neither in your Kastel nor in a working copy.
That mechanism still holds when Deel changes. If the publisher adds a sensitive field to a worker's record tomorrow, that field will arrive in the response and leave again without being written, because the connector never reads anything beyond its four fields. A record of an unexpected shape produces a shorter note, never a wider one.
A staff directory does not get broadcast to the rest of the company
Deel sets no read permission at the level of a person. The organisation key sees the entire directory, and it is the same key for every record. There is therefore no original permission to inherit, unlike a discussion channel or a shared file.
The connector then files the whole directory in a single department, and only if three conditions hold. An administrator must have designated the department that runs human resources, that name must be a valid identifier, and that department must genuinely exist in your organisation. If any one of the three is missing, the directory stays on hold and nobody reads it.
One choice deserves naming, because the opposite would have been easier to write. The connector never files a person's record in the department where that person works. That would be the natural reflex, and the result would be a staff directory scattered across every department, each one reading the records of its own members. On staff data, the connector would rather wait than land there.
This way of deciding who reads what is not specific to Deel. It is how your Kastel governs all of its content, source by source.
The fields that come in, and the ones that stay out
A directory is judged on its fields, so here are the two exact lists the connector applies to every record.
These exclusions are not settings a console could reopen. They follow from the connector reading only the four fields in the first list, and a field that is not read cannot be written.
The questions an HR director always asks
Can an AI reach a salary through this connector?
No. The connector names the four fields it writes, and pay is not one of them. The field is never read, so there is no filter to get around. The same goes for the date of birth, the personal email address, the phone numbers, the identity document and the record's additional details.
Who can query the directory once it is in place?
The people in the department an administrator designated, and nobody else. An AI working for someone in another department does not reach these records. Until a department is designated, the directory waits, and no AI reads it whatever account is used.
What happens to the record of someone who leaves the company?
Deel keeps departed people in its directory, with a status flag. For the connector, a departure is therefore an edit to the record rather than a disappearance, and the record is reread as it now stands. When a person is genuinely removed from Deel's directory, their record is marked in your Kastel as gone at the source and kept, readable by the same people as before, until an administrator decides to erase it.
Does Kastel read the department list in your Deel account?
No. A person's department is read off their own record, as Deel writes it. The connector does not query your account's department list, and it does not try to reconstruct your official org chart anywhere other than in the records themselves.
What this connector does not do
It does not replace your HR system. It manages no leave, prepares no payroll, tracks no contract and writes nothing into Deel. The authorisation requested is a read authorisation, there is no write path to your account, and the connector talks to nothing but Deel's API domain.
The org chart it gives an AI is worth exactly what your Deel directory is worth. If a person's department is blank, or their manager changed without the record following, your Kastel carries the error over as it stands. The connector corrects nothing and invents no reporting line.
On this source there is no line-by-line sorting of content, as there is on a mailbox. The protection rests entirely on the four fields read and on the single department the directory is filed in. A person's name is still personal data, and the connector writes one per record, which is the very reason nothing is readable before a department has been designated.
Every pass rereads the whole directory, because Deel cannot say which records have changed since last time. A new hire therefore shows up on the following pass rather than the minute the record is created. And when a page of results announces more to come without giving the means to reach it, the pass fails loudly instead of wrongly concluding that the unread records have gone.
Before you connect your staff directory, let us talk.
An HR directory is better discussed between two people than through a web page. Tell us which questions you want an AI to handle about your organisation, and we will tell you what this connector reads, what it leaves out, and where it stops.
Get in touch
